Hi there,
Since the entire corpus should be based on features like user's interest, their behaviour etc., I feel we can approach in following ways:
1. TF-IDF could be used to prepare matrix that influences interest of a user to a greater extent for a particular post, pproduct, profile etc.
2. We can also use Sentiment/Emotional Analysis in order to understand user's behaviour (preferably using Vader Algo, since its been trained on social media data)
As far as your second point is concerned I would need more clarity on that
I would be really interested on discussing further about the project.
Thanks!!